The Direct3D 10 and higher pipeline contains three programmable-shader stages (the rounded blocks in the pipeline functional diagram).
Each of these shader stages exposes its own unique functionality, built on the shader model 4.0 common-shader core.
The vertex-shader (VS) stage processes vertices from the input assembler, performing per-vertex operations such as transformations, skinning, morphing, and per-vertex lighting. Vertex shaders always operate on a single input vertex and produce a single output vertex. The vertex shader stage must always be active for the pipeline to execute. If no vertex modification or transformation is required, a pass-through vertex shader must be created and set to the pipeline.
Each vertex shader input vertex can be comprised of up to 16 32-bit vectors (up to 4 components each) and each output vertex can be comprised of as many as 16 32-bit 4-component vectors. All vertex shaders must have a minimum of one input and one output, which can be as little as one scalar value.
The vertex-shader stage can consume two system generated values from the input assembler: VertexID and InstanceID (see System Values and Semantics). Since VertexID and InstanceID are both meaningful at a vertex level, and IDs generated by hardware can only be fed into the first stage that understands them, these ID values can only be fed into the vertex-shader stage.
Vertex shaders are always run on all vertices, including adjacent vertices in input primitive topologies with adjacency. The number of times that the vertex shader has been executed can be queried from the CPU using the VSInvocations pipeline statistic.
A vertex shader can perform load and texture sampling operations where screen-space derivatives are not required (using HLSL intrinsic functions: Sample (DirectX HLSL Texture Object), SampleCmpLevelZero (DirectX HLSL Texture Object), and SampleGrad (DirectX HLSL Texture Object)).
The geometry-shader (GS) stage runs application-specified shader code with vertices as input and the ability to generate vertices on output. Unlike vertex shaders, which operate on a single vertex, the geometry shader's inputs are the vertices for a full primitive (two vertices for lines, three vertices for triangles, or single vertex for point). Geometry shaders can also bring in the vertex data for the edge-adjacent primitives as input (an additional two vertices for a line, an additional three for a triangle). The following illustration shows a triangle and a line with adjacent vertices.
The geometry-shader stage can consume the SV_PrimitiveID system-generated value that is auto-generated by the IA. This allows per-primitive data to be fetched or computed if desired.
The geometry-shader stage is capable of outputting multiple vertices forming a single selected topology (GS stage output topologies available are: tristrip, linestrip, and pointlist). The number of primitives emitted can vary freely within any invocation of the geometry shader, though the maximum number of vertices that could be emitted must be declared statically. Strip lengths emitted from a geometry shader invocation can be arbitrary, and new strips can be created via the RestartStrip HLSL function.
Geometry shader output may be fed to the rasterizer stage and/or to a vertex buffer in memory via the stream output stage. Output fed to memory is expanded to individual point/line/triangle lists (exactly as they would be passed to the rasterizer).
When a geometry shader is active, it is invoked once for every primitive passed down or generated earlier in the pipeline. Each invocation of the geometry shader sees as input the data for the invoking primitive, whether that is a single point, a single line, or a single triangle. A triangle strip from earlier in the pipeline would result in an invocation of the geometry shader for each individual triangle in the strip (as if the strip were expanded out into a triangle list). All the input data for each vertex in the individual primitive is available (i.e. 3 vertices for triangle), plus adjacent vertex data if applicable/available.
A geometry shader outputs data one vertex at a time by appending vertices to an output stream object. The topology of the streams is determined by a fixed declaration, choosing one of: PointStream, LineStream, or TriangleStream as the output for the GS stage. There are three types of stream objects available, PointStream, LineStream and TriangleStream which are all templated objects. The topology of the output is determined by their respective object type, while the format of the vertices appended to the stream is determined by the template type. Execution of a geometry shader instance is atomic from other invocations, except that data added to the streams is serial. The outputs of a given invocation of a geometry shader are independent of other invocations (though ordering is respected). A geometry shader generating triangle strips will start a new strip on every invocation.
When a geometry shader output is identified as a System Interpreted Value (e.g. SV_RenderTargetArrayIndex or SV_Position), hardware looks at this data and performs some behavior dependent on the value, in addition to being able to pass the data itself to the next shader stage for input. When such data output from the geometry shader has meaning to the hardware on a per-primitive basis (such as SV_RenderTargetArrayIndex or SV_ViewportArrayIndex), rather than on a per-vertex basis (such as SV_ClipDistance[n] or SV_Position), the per-primitive data is taken from the leading vertex emitted for the primitive.
Partially completed primitives could be generated by the geometry shader if the geometry shader ends and the primitive is incomplete. Incomplete primitives are silently discarded. This is similar to the way the IA treats partially completed primitives.
The geometry shader can perform load and texture sampling operations where screen-space derivatives are not required (samplelevel, samplecmplevelzero, samplegrad).
Algorithms that can be implemented in the geometry shader include:
- Point Sprite Expansion
- Dynamic Particle Systems
- Fur/Fin Generation
- Shadow Volume Generation
- Single Pass Render-to-Cubemap
- Per-Primitive Material Swapping
- Per-Primitive Material Setup - Including generation of barycentric coordinates as primitive data so that a pixel shader can perform custom attribute interpolation (for an example of higher-order normal interpolation, see CubeMapGS Sample).
The pixel-shader stage (PS) enables rich shading techniques such as per-pixel lighting and post-processing. A pixel shader is a program that combines constant variables, texture data, interpolated per-vertex values, and other data to produce per-pixel outputs. The rasterizer stage invokes a pixel shader once for each pixel covered by a primitive, however, it is possible to specify aNULL shader to avoid running a shader.
When multisampling a texture, a pixel shader is invoked once per-covered pixel while a depth/stencil test occurs for each covered multisample. Samples that pass the depth/stencil test are updated with the pixel shader output color.
The pixel shader intrinsic functions produce or use derivatives of quantities with respect to screen space x and y. The most common use for derivatives is to compute level-of-detail calculations for texture sampling and in the case of anisotropic filtering, selecting samples along the axis of anisotropy. Typically, a hardware implementation runs a pixel shader on multiple pixels (for example a 2x2 grid) simultaneously, so that derivatives of quantities computed in the pixel shader can be reasonably approximated as deltas of the values at the same point of execution in adjacent pixels.
When the pipeline is configured without a geometry shader, a pixel shader is limited to 16, 32-bit, 4-component inputs. Otherwise, a pixel shader can take up to 32, 32-bit, 4-component inputs.
Pixel shader input data includes vertex attributes (that can be interpolated with or without perspective correction) or can be treated as per-primitive constants. Pixel shader inputs are interpolated from the vertex attributes of the primitive being rasterized, based on the interpolation mode declared. If a primitive gets clipped before rasterization, the interpolation mode is honored during the clipping process as well.
Vertex attributes are interpolated (or evaluated) at pixel shader center locations. Pixel shader attribute interpolation modes are declared in an input register declaration, on a per-element basis in either an argument or an input structure. Attributes can be interpolated linearly, or with centroid sampling. Centroid evaluation is relevant only during multisampling to cover cases where a pixel is covered by a primitive but a pixel center may not be; centroid evaluation occurs as close as possible to the (non-covered) pixel center.
Inputs may also be declared with a system-value semantic, which marks a parameter that is consumed by other pipeline stages. For instance, a pixel position should be marked with the SV_Position semantic. The IA stage can produce one scalar for a pixel shader (using SV_PrimitiveID); the rasterizer stage can also generate one scalar for a pixel shader (using SV_IsFrontFace).
A pixel shader can output up to 8, 32-bit, 4-component colors, or no color if the pixel is discarded. Pixel shader output register components must be declared before they can be used; each register is allowed a distinct output-write mask.
Use the depth-write-enable state (in the output-merger stage) to control whether depth data gets written to a depth buffer (or use the discard instruction to discard data for that pixel). A pixel shader can also output an optional 32-bit, 1-component, floating-point, depth value for depth testing (using the SV_Depth semantic). The depth value is output in the oDepth register, and replaces the interpolated depth value for depth testing (assuming depth testing is enabled). There is no way to dynamically change between using fixed-function depth or shader oDepth.
A pixel shader cannot output a stencil value.